Institute of Energy Technologies (IET)

Grundlagen der Elektrochemie (IET-1)

CECAM Flagship Workshop: Machine Learning of First Principles Observables

We thank all the participants for their contributions!

Overview and Scientific Program

The workshop on “Machine learning of first principles observables” was aimed at addressing the growing need for models, workflows, and databases which go beyond the established methods of producing machine learning (ML) inter-atomic potentials, and instead serve to predict experimentally observable quantities. During the event, eight subject-specific sessions aimed at bridging these experimental and theory communities, from “Thermodynamic observables” to “Long range interactions” and “Spectroscopy”, where speakers both from theory and experiment were invited. Additionally, a panel discussion after each session of talks addressed how that specific field was advancing with respect to ML for observables.

A poster session allowed many participants to present their work. The Best Poster Awards were awarded to Patricia König (Fritz-Haber-Institut, 1st), Pol Febrer Calabozo (ICN2, 2nd), and Marvin Friede (University of Bonn, 3rd). Non-scientific content in form of coffee breaks, a conference dinner, and an outing with a Berlin city tour enabled the participants to discuss and connect in a less formal setting.

Workshop Program

Session and Discussion Topics

The eight sessions of the workshop were focusing on the following topics:

Thermodynamic Observables
Electronic Structure and Long Range Interactions (3 sessions)
Magnetic Observables
Spectroscopic Observables (2 sessions)
Databases and Reaction Networks

Overarching all sessions, there were several topics that were identified to be very important in forming this community:

Data sharing and management: In almost every panel discussion, the importance of effective data sharing, meta-data utilization, and the creation and maintenance of curated databases was discussed. It was also emphasized that these databases should include negative results, which aid ML models in learning what not to do and are critical for experimentally relevant ML models for the future. Code documentation and reproducibility was also a feature of these discussions.

Bridging Experiment and Simulation: The workshop itself served as a springboard to facilitate exchange between theoreticians and experimentalists. By encouraging discussion between both groups, the speakers and participants identified several areas where these two groups could bridge the multi-scaling gap from both ends. This involves theoreticians reconsidering the approximations and simplifications in their models to make them more realistic by incorporating factors such as interfaces and defects. At the same time, experimentalists were encouraged to conduct simplified, idealized benchmarking experiments. This dual approach aims to bring theory and experiment closer together, bridging the complexity gap from both ends.

Metrics for Evaluating Predicted Data: The final topic discussed at the event was the need for better metrics for evaluating the accuracy of predicted data beyond simple scalar values. It included discussing metrics which allow for tolerance in variations in spectra shifts, peak width, and spectral intensities. Additionally, the importance of foundational ML models such as MACE-MP-0, ChargeNET, and AIMNet2 models was highlighted.

By addressing these topics, the workshop has not only pushed the boundaries of current methodologies but also fostered a collaborative environment that bridges theoretical and experimental approaches, ultimately contributing to more robust and applicable scientific advancements. Furthermore, there were several tangible outcomes of the workshop, especially with regards to the dissemination of open source codes, new collaboration between experimentalists and theorists, and discussion of new publications combining the research of different participants.

In summation, the workshop successfully identified the critical areas for improvement within the fields of predicting observables from first principles, and established a network for future research efforts.